Driving Data & AI Product Leadership.
Building intelligent systems that turn complex data into actionable insights—from agentic AI platforms to research-backed analytics.
Professional Portfolio
AI Data Analyst: Databricks Natural Language Query Platform
An enterprise self-service analytics platform that translates conversational questions (in plain English) into accurate SQL database queries via the Databricks Genie API. It enables business teams and product managers to analyze buyer journeys, inspect account histories, and enrich large customer datasets without needing technical database coding skills.
- Conversational Queries: Ask business questions in natural English to instantly query enterprise database tables.
- Buyer Journey Insights: Look up account history, interaction timelines, and product purchases by member ID or contact.
- Smart Entity Matching: Connects records across multi-workspace data using emails, organizational hierarchies, and account IDs.
- Bulk Data Enrichment: Automatically enriches thousands of spreadsheet rows with live database attributes at scale.
Voice of Customer (VOC) & Buyer Journey AI Assistant
A production-ready customer intelligence assistant built with a "dual-brain" architecture. It combines structured database records (orders, timeline logs) with unstructured documents (customer feedback PDFs, interview transcripts), automatically routing user questions to the right data source to deliver unified customer insights.
- Smart Dual-Source Routing: Automatically routes questions to database queries (SQL) or document libraries (PDF search).
- Workflow Orchestration: Uses LangGraph to manage multi-step reasoning, conversation memory, and error recovery.
- Semantic Document Search: Finds relevant feedback excerpts using vector embeddings and similarity search.
- Production Observability: Monitored with LangSmith for response accuracy, latency, and query safety.
Automated A/B Testing & Marketing Experiment Analyzer
An interactive analytics platform that evaluates marketing experiment results with statistical rigor. It automatically runs hypothesis tests, checks sample size confidence (statistical power), identifies winning campaign variations, and translates mathematical outcomes into plain-English recommendations for growth teams.
- Rigorous Hypothesis Testing: Automates 4 standard statistical tests (Z-test, Chi-Square, Binomial, Power Analysis).
- Trustworthy Winner Calls: Enforces 80%+ statistical power thresholds to prevent declaring false winners prematurely.
- Plain-English Explanations: Combines LLMs with statistical results to generate executive-ready summaries.
- Subgroup & Segment Breakdown: Analyzes how different user segments responded to each experiment variation.
Automated LinkedIn Ad Measurement & Account Intelligence Pipeline
An automated data pipeline that connects directly to the LinkedIn Marketing API to extract unsampled, daily ad performance data. It enriches raw campaign logs with verified company names and generates automated reporting feeds for marketing leaders and cross-platform attribution models.
- Full Daily Granularity: Extracts exact daily campaign metrics without UI estimation or data sampling.
- Account & Company Enrichment: Automatically maps raw account IDs to real company names via API lookup.
- Automated Stakeholder Reporting: Generates formatted executive Excel and CSV reports on a scheduled basis.
- Cross-Platform Attribution Feeds: Feeds clean LinkedIn advertising data into downstream measurement systems.
Google Analytics (GA4) Raw Data Pipeline & Unsampled Reporting
A high-volume data pipeline that extracts raw web traffic and conversion events directly via the Google Analytics 4 (GA4) API. It bypasses standard web interface limits and data sampling, exporting 100,000+ rows of detailed user interaction data straight into data warehouses for accurate attribution.
- Zero Data Sampling: Retrieves full, un-sampled datasets for 100% accurate conversion and traffic analysis.
- High-Volume Extraction: Bypasses UI row limits to export 100,000+ event rows per batch.
- Complete Event Parameters: Captures custom user dimensions, campaign UTM parameters, and conversion milestones.
- Warehouse Integration: Loads structured event data directly into cloud warehouses (BigQuery/Snowflake) for BI.
Personal Portfolio
OpenClaw: Autonomous WhatsApp & Voice Personal Assistant
An autonomous everyday assistant connected via WhatsApp and local voice control (running on a private Raspberry Pi). It automates personal productivity by reading calendar events, summarizing high-priority emails, monitoring financial news, and delivering structured morning briefings.
- Daily Morning Briefings: Automatically digests emails, calendar events, and market news into concise daily summaries.
- Multi-Modal Interaction: Communicates via WhatsApp messaging and voice commands.
- Privacy-First Local Execution: Runs private AI models on local hardware (Raspberry Pi) for security and low latency.
- Scheduled Autonomous Actions: Executes background workflows like reminder notifications and news aggregation on autopilot.
Infinite Memory: Solving Long-Term AI Agent Recall (Research Paper)
A published research paper solving a critical failure in AI agents: traditional AI memory collapses (recalling only ~17% of key facts) over extended conversations. This paper introduces Recursive Gated Consolidation (RGC), a hierarchical memory architecture that preserves 100% fact recall across 10 million conversation turns with predictable, constant-time speed.
- Identified Memory Collapse ("Discovery Cliff"): Documented how standard AI models lose recall over long conversation histories.
- Recursive Memory Compression (RGC): Condenses background history into layered memory tiers without losing key facts.
- Constant-Speed Performance: Ensures the AI searches memories instantly regardless of whether the conversation is 10 turns or 10 million turns long.
- Rigorous Cross-Model Validation: Empirically benchmarked across frontier models (Google Gemini 2.5 to 3.1) and hardware architectures.
Personal AI Infrastructure (PAI v2): Persistent AI Developer System
A modular foundation layer that turns AI coding assistants into persistent engineering partners. It retains project memory, architectural decisions, and error solutions across coding sessions, eliminating repetitive explanations and preventing recurrent bugs across all local repositories.
- Persistent Long-Term Memory: Automatically records architectural decisions and bug fixes so AI assistants never forget past context.
- Standardized Quality Cycles: Enforces a 6-step engineering discipline (Observe, Think, Plan, Build, Execute, Verify).
- Cross-Project Learning: Shares hardened solutions and best practices across multiple independent software codebases.
- Session & API Resilience: Automatically prevents API rate-limit loops and token overflow crashes during intensive coding.
Hierarchical Multi-Agent AI Coordination System
An enterprise-grade multi-agent architecture inspired by cognitive science ("System 1 / System 2" thinking). It separates high-level strategic planning from tactical code execution, using a parent supervisor AI to enforce safety checks while specialized worker sub-agents complete tasks in parallel.
- Strategic Supervisor (Parent Agent): Defines requirements, manages task sequencing, and verifies output quality before approval.
- Specialized Worker Agents (Sub-Agents): Execute focused tasks (coding, testing, documentation) at maximum velocity.
- Automated Quality Gates: Validates every intermediate step with automated tests before passing work forward.
- Prevents Agent Drift: Keeps autonomous multi-step tasks aligned with the original product goals.
Autonomous Stock & Crypto Trading Engine (V4.1)
A quantitative investment engine that autonomously analyzes market data and manages a diversified stock and crypto portfolio. The algorithm blends short-term trend momentum with fundamental financial metrics, validated by real-time paper portfolio tracking and disciplined risk controls.
- High-Performance Strategy: Demonstrates 80.8% annualized return (XIRR), outperforming the S&P 500 benchmark in backtests and paper runs.
- Dual-Signal Engine: Combines technical momentum indicators (e.g., trend crossovers) with company balance sheet fundamentals.
- Real-Time Signal Verification ("Shadow Tracking"): Validates algorithm predictions against live market data before execution.
- Tax-Aware Portfolio Rebalancing: Prioritizes long-term capital gains holding periods to maximize net post-tax returns.
Universal AI Agent Orchestration & Safety Framework
A cross-platform developer tool that standardizes AI agent workflows across any code editor (VS Code, JetBrains, or Terminal CLI). It acts as a safety supervisor, enforcing automatic file protection, runtime locks, and circuit breakers that prevent AI agents from getting stuck in runaway loops.
- Works Across Any Editor: Standardizes agent commands seamlessly across VS Code, PyCharm, and command-line interfaces.
- Automated Circuit Breakers: Detects and halts repetitive error loops before they waste API tokens or compute resources.
- Safe File Modifications: Guards critical repository files against unintended bulk overwrites or corrupt edits.
- Policy-Based Execution: Enforces mandatory test verification before any code changes are marked complete.
Planetary & Astrological Data Intelligence Platform
A full-stack web application that combines high-precision planetary calculations with Generative AI interpretations. It evaluates historical biographical datasets across thousands of individuals to test astrological patterns using rigorous statistical validation methods.
- Large-Scale Data Analysis: Evaluated planetary configurations across 5,800+ historical profiles and 200+ geographic regions.
- Statistical Validation: Applied False Discovery Rate (FDR) statistical corrections to separate true correlation from random noise.
- Precision Planetary Math: Computes high-accuracy celestial coordinates using astronomical ephemeris engines.
- Modern Full-Stack Architecture: Built with Next.js (React) on the frontend, FastAPI (Python) on the backend, and PostgreSQL.
Philosophy Sage: Knowledge Graph & Conversational AI for Classical Literature
An interactive conversational AI platform that allows users to explore ancient philosophical literature (like the Bhagavad Gita and Mahabharata). By linking character relationships, themes, and events into a connected Knowledge Graph, it accurately answers nuanced questions and plays original Sanskrit verse audio.
- Knowledge Graph + Vector Search (GraphRAG): Maps complex character relationships and historical lineages for deep contextual answers.
- Graph Database Architecture: Indexes thousands of interconnected entities across 18 epic books in Neo4j.
- Audio Verse Playback: Features Text-to-Speech (TTS) integration that reads original Sanskrit verses aloud.
- Conversational Question-Answering: Built with LangChain and React for smooth, interactive exploration.
Sonic Geometry: Real-Time Audio & Harmonic Waveform Visualizer
An interactive web application that visualizes the physics and mathematics of sound in real time. Built entirely in vanilla JavaScript with the Web Audio API, users can play musical notes on a virtual keyboard and instantly see frequency spectrums, sound waves (oscilloscope), and harmonic geometric patterns.
- Real-Time Sound Waveforms: Displays live frequency spectrums (FFT) and oscilloscope waveforms as notes are played.
- Harmonic Geometry (Lissajous Curves): Visualizes musical intervals and chord ratios as geometric wave patterns.
- Interactive Music Theory Tools: Includes an interactive Circle of Fifths and playable polyphonic synthesizer keyboard.
- Ultra-Fast & Lightweight: Built in pure vanilla JavaScript with zero external frameworks or heavyweight dependencies.
Technical Stack
Data Foundations
API-driven pipelines, Event Taxonomies, Schema Design, Data Monetization Strategy, Bulk Enrichment.
Ad Tech & Measurement
GA4, Google Ads, LinkedIn Ads, Conversion Modeling, Cross-Platform Attribution, A/B Statistical Testing.
Agentic & GenAI
Databricks Genie, LangGraph, LangChain, RAG Architectures, Text-to-SQL, Vector Search, LLM Integration.